wltr

I won’t place a 7th generation Intel CPU as slow and outdated. You may be surprised to learn I have multiple Intel Atom 230/330 CPUs that serve my personal needs very well. It’s extremely slow CPU, yet it’s surprisingly capable for a home server setup.

To give you some context: a 4K JPEG image is converted to AVIF within 30 seconds on my i7-4770 desktop, while said Intel Atom processor does the same job within 30 minutes! I don’t expect Immich to run all the processing on such machine, obviously. But that would be cool to host a web site with the archives on said hardware (it’s plenty capable to run a simple web server, especially for just one user / or a family).

Also, I don’t understand the scale model. I’d rather make a reverse proxy and a bunch of old computers like that, so when I need to access a photo from, say, 2010, it would wake on LAN start the associated computer, instead of me needing to have a super beefy computer with 20 TB disks that does everything. That’s why I think the architecture does not have the scaling in mind, or it’s not obvious / trivial.

My current Immich server is the MacBook Pro from 2011, mobile i5 something (3rd generation I believe). It’s OK, when things are processed. But I’d prefer to offload it to ultra-low-powered machines for scale and power efficiency. Again: perhaps that’s possible already, I haven’t investigated it thoroughly yet.

I have this impression because of some initially different decisions to a self-hosted architecture (where you usually have not that powerful machines at home), it feels like a proprietary original service they replicate, which is designed to serve millions of users off beefy machines. And because of how rapid the development is. I’d be surprised to learn it’s not heavily vibe coded, considering all the aspects of the project.

I did not disable all the features, because I had to pick a beefy server for it, otherwise I’d started using the project years ago (I’m only months in). I confirm, it was super slow first days (I have about 500 GB library), then it settled down. I’d love it to be designed around comfortably living on a cheap weak hardware, as I believe just accessing my library (and not processing it) should be trivial. I just hope against hope that would be possible at some later stage of the development. Perhaps that is possibly today, I haven’t investigated too much. But I guess the documentation would cover it then.

I’d love these two to be separated, so I could utilise my powerful PC to process everything, and then it would live on the NAS / SBC / anything smol.

Yeah, I agree. That’s an interesting phenomenon.

I haven’t investigated, that’s my personal impression based on the architecture I think wasn’t given much thought (self-hosting and proprietary service for millions users with expensive hardware are very different) and how hungry it is for resources. I’d love to be very wrong here. Overall, despite me thinking it’s heavily vibe coded, I still make an exception for it and use it and recommend it, as it’s useful and truly easy to use. I just accepted the fact I need a beefy computer to host, instead of processing it on a beefy computer and then hosting on a NAS.

I’m using Immich now. I’m not happy with how hunger it is for the resources, it feels heavily vibe coded. Yet, it’s quite a clone of Google Photos, and not a bad one (interface and UX wise), so I just hope it’ll be optimised at some later time.